I/O Characterization of Big Data Workloads in Data Centers

As the amount of data explodes rapidly, more and more organizations tend to use data centers to make effective decisions and gain a competitive edge. Big data applications have gradually dominated the data centers’ workloads, and hence it has been increasingly important to understand their behaviour in order to further improve the performance of data centers. Due to the constantly increased gap between I/O devices and CPUs, I/O performance dominates the overall system performance, so characterizing I/O behaviour of big data workloads is important and imperative. In this paper, we select four typical big data workloads in broader areas from the BigDataBench which is a big data bench-mark suite from internet services. They are Aggregation, TeraSort, Kmeans and PageRank. We conduct detailed deep analysis of their I/O characteristics, including disk read/write bandwidth, I/O devices’ utilization, average waiting time of I/O requests, and average size of I/O requests, which act as a guide to design highperformance, low-power and cost-aware big data storage systems.